diff --git a/CHANGELOG.md b/CHANGELOG.md
index 4e33026a58..4125838100 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,6 +4,7 @@ This file contains the changelog for the Deeploy project. The changelog is divid
 ## Unreleased (Planned Release Target: v0.2.1)
 
 ### List of Pull Requests
+- Improve Profiling [#138](https://github.com/pulp-platform/Deeploy/pull/138)
 - FP32 ReduceMean operator improvement [#137](https://github.com/pulp-platform/Deeploy/pull/137)
 - Support for RMSNorm (Pow and Sqrt operators) [#136](https://github.com/pulp-platform/Deeploy/pull/136)
 - Demo TinyViT compatibility with tiled Siracusa [#124](https://github.com/pulp-platform/Deeploy/pull/124)
@@ -81,6 +82,7 @@ This file contains the changelog for the Deeploy project. The changelog is divid
 - Added new waiting-strategy logic with fine-grained `PerTensorWaitingStrategy`
 - PULPClusterEngine now accepts a `n_cores` parameter to set the number of cores used
 - annotateNCores method to PULPDeployer that adds an `n_cores` key to all PULPClusterEngine templates' operatorRepresentations
+- Calculate non-kernel overhead and show total time spent during profiling
 
 ### Changed
 - Structure of Tests subdir for improved ordering
@@ -123,6 +125,7 @@ This file contains the changelog for the Deeploy project. The changelog is divid
 - Added missing shape annotation to the testTypeInferenceDifferentTypes
 - Refactored DMA code generation (`SnitchDma`, `Mchan`) to correctly overlap transfers and compute in double-buffering mode
 - changed `_mapNode` to `_selectEngine` which reduces the responsibility of that function to, as the name states, just engine selection
+- Print kernel profiling information for all memory levels
 
 ### Fixed
 - Fixed ReduceMean parallelization and tiling issues described in Issue [#134](https://github.com/pulp-platform/Deeploy/issues/134).
diff --git a/Deeploy/TilingExtension/CodeTransformationPasses/DoubleBufferingTilingCodeGeneration.py b/Deeploy/TilingExtension/CodeTransformationPasses/DoubleBufferingTilingCodeGeneration.py
index d436d1ccc2..ad9c6ad012 100644
--- a/Deeploy/TilingExtension/CodeTransformationPasses/DoubleBufferingTilingCodeGeneration.py
+++ b/Deeploy/TilingExtension/CodeTransformationPasses/DoubleBufferingTilingCodeGeneration.py
@@ -276,17 +276,13 @@ def _tilingLoop(self, ctxt: NetworkContext, executionBlock: ExecutionBlock,
         teardownStatements.append(CodeSnippet(self._lineComment, {"comment": "Deinitialize DMA future"}))
         teardownStatements.extend(f.deinit() for f in ingressFutures | egressFutures)
 
-        metaInfo = TilingMetaInfo(
-            nodeName = operatorRepresentation['nodeName'] + f"_{self.externalMemory}",
-            nodeOps = operatorRepresentation['nodeOps'],
-            numTiles = operatorRepresentation['numTiles'],
-            totalNumTiles = len(tilingSchedule.outputLoadSchedule),
-            tileIdxPtr = operatorRepresentation['tileIdxPtr'],
-            tileIdxVar = "TILING_I",
-            # TODO: The kernelLevelTiling field is used in profiling to know we are generating code around the kernel.
-            #       The current implementation does this by checking whether we are at the lowest memory level,
-            #       which is hardcoded by the value "L1". Change this to be memory level agnostic.
-            kernelLevelTiling = self.localMemory == "L1")
+        metaInfo = TilingMetaInfo(nodeName = operatorRepresentation['nodeName'] + f"_{self.externalMemory}",
+                                  nodeOps = operatorRepresentation['nodeOps'],
+                                  numTiles = operatorRepresentation['numTiles'],
+                                  totalNumTiles = len(tilingSchedule.outputLoadSchedule),
+                                  tileIdxPtr = operatorRepresentation['tileIdxPtr'],
+                                  tileIdxVar = "TILING_I",
+                                  kernelLevelTiling = True)
 
         executionBlock = self.generateAllTilingCode(executionBlock, metaInfo, ingressDMAStatements, egressDMAStatements,
                                                     openLoopStatements, closeLoopStatements, setupStatements,
diff --git a/Deeploy/TilingExtension/CodeTransformationPasses/SingleBufferingTilingCodeGeneration.py b/Deeploy/TilingExtension/CodeTransformationPasses/SingleBufferingTilingCodeGeneration.py
index fb08a0e818..ea1e938b58 100644
--- a/Deeploy/TilingExtension/CodeTransformationPasses/SingleBufferingTilingCodeGeneration.py
+++ b/Deeploy/TilingExtension/CodeTransformationPasses/SingleBufferingTilingCodeGeneration.py
@@ -117,17 +117,13 @@ def _tilingLoop(self, ctxt: NetworkContext, executionBlock: ExecutionBlock,
 
         closeLoopStatements = [CodeSnippet(self._closeTileLoopTemplate, {**operatorRepresentation})]
 
-        metaInfo = TilingMetaInfo(
-            nodeName = operatorRepresentation['nodeName'] + f"_{self.externalMemory}",
-            nodeOps = operatorRepresentation['nodeOps'],
-            numTiles = operatorRepresentation['numTiles'],
-            totalNumTiles = len(tilingSchedule.outputLoadSchedule),
-            tileIdxPtr = operatorRepresentation['tileIdxPtr'],
-            tileIdxVar = "TILING_I",
-            # TODO: The kernelLevelTiling field is used in profiling to know we are generating code around the kernel.
-            #       The current implementation does this by checking whether we are at the lowest memory level,
-            #       which is hardcoded by the value "L1". Change this to be memory level agnostic.
-            kernelLevelTiling = self.localMemory == "L1")
+        metaInfo = TilingMetaInfo(nodeName = operatorRepresentation['nodeName'] + f"_{self.externalMemory}",
+                                  nodeOps = operatorRepresentation['nodeOps'],
+                                  numTiles = operatorRepresentation['numTiles'],
+                                  totalNumTiles = len(tilingSchedule.outputLoadSchedule),
+                                  tileIdxPtr = operatorRepresentation['tileIdxPtr'],
+                                  tileIdxVar = "TILING_I",
+                                  kernelLevelTiling = True)
 
         executionBlock = self.generateAllTilingCode(executionBlock, metaInfo, ingressDMAStatements, egressDMAStatements,
                                                     openLoopStatements, closeLoopStatements, setupStatements,
diff --git a/Deeploy/TilingExtension/CodeTransformationPasses/TilingPrototypes.py b/Deeploy/TilingExtension/CodeTransformationPasses/TilingPrototypes.py
index 2f6c1e9590..09a4ef56eb 100644
--- a/Deeploy/TilingExtension/CodeTransformationPasses/TilingPrototypes.py
+++ b/Deeploy/TilingExtension/CodeTransformationPasses/TilingPrototypes.py
@@ -79,14 +79,27 @@ class ProfilingPrototypeMixIn(ABC):
 
     _printLoopSetup = NodeTemplate("""
     StopTimer();
+    printf("===== Profiling ${nodeName} =====\\n");
     for (int ${profileIdxVar} = ((*${tileIdxPtr} > 0) ? ${numTiles}[(*${tileIdxPtr} - 1)] : 0);
         ${profileIdxVar} < ${numTiles}[*${tileIdxPtr}];
         ${profileIdxVar}++){
     """)
 
-    _printCycleDifference = NodeTemplate(r"""
-    printf("%s%u] %s%u%s", ${prefixStr}, ${profileIdxVar}, "${flavorStr}", \
-    ${measurementsEnd}[${profileIdxVar}] - ${measurementsStart}[${profileIdxVar}], ${suffixStr});
+    _measurementDeclaration = NodeTemplate("""
+    uint32_t ${measurement} = ${measurementsEnd}[${profileIdxVar}] - ${measurementsStart}[${profileIdxVar}];
+    """)
+
+    _printCycleDifference = NodeTemplate("""
+    printf("%s%u] %s%6u%s", ${prefixStr}, ${profileIdxVar}, "${flavorStr}", \
+    ${measurement}, ${suffixStr});
+    """)
+
+    _printCycleContribution = NodeTemplate("""
+    uint32_t total = ${measurementInput} + ${measurementKernel} + ${measurementOutput};
+    uint32_t dma = ${measurementInput} + ${measurementOutput};
+    float overhead_percentage = (total == 0) ? 0 : dma * 100.0f / total;
+    float kernel_percentage = (total == 0) ? 0 : ${measurementKernel} * 100.0f / total;
+    printf("%s%u] Total      :%6u cycles (%2.1f%% Kernel + %2.1f%% Overhead, %u + %u)\\n", ${prefixStr}, ${profileIdxVar}, total, kernel_percentage, overhead_percentage    , ${measurementKernel}, dma);
     """)
 
     _printLoopTeardown = NodeTemplate("""
@@ -151,13 +164,37 @@ def injectPrintCycleDiff(cls, executionBlock: ExecutionBlock, metaInfo: TilingMe
             "tileIdxPtr": tileIdxPtr,
         })
 
+        executionBlock.addRight(
+            cls._measurementDeclaration, {
+                "measurement": f"{nodeName}_ingress_dma_wait_measurement",
+                "measurementsStart": f"{nodeName}_ingress_dma_wait_start_measurements",
+                "measurementsEnd": f"{nodeName}_ingress_dma_wait_end_measurements",
+                "profileIdxVar": profileIdxVar,
+            })
+
+        if metaInfo.kernelLevelTiling:
+            executionBlock.addRight(
+                cls._measurementDeclaration, {
+                    "measurement": f"{nodeName}_kernel_measurement",
+                    "measurementsStart": f"{nodeName}_kernel_start_measurements",
+                    "measurementsEnd": f"{nodeName}_kernel_end_measurements",
+                    "profileIdxVar": profileIdxVar,
+                })
+
+        executionBlock.addRight(
+            cls._measurementDeclaration, {
+                "measurement": f"{nodeName}_egress_dma_wait_measurement",
+                "measurementsStart": f"{nodeName}_egress_dma_wait_start_measurements",
+                "measurementsEnd": f"{nodeName}_egress_dma_wait_end_measurements",
+                "profileIdxVar": profileIdxVar,
+            })
+
         executionBlock.addRight(
             cls._printCycleDifference, {
                 "prefixStr": f"{nodeName}_prefix",
                 "suffixStr": f"{nodeName}_suffix",
-                "flavorStr": "Input DMA took ",
-                "measurementsStart": f"{nodeName}_ingress_dma_wait_start_measurements",
-                "measurementsEnd": f"{nodeName}_ingress_dma_wait_end_measurements",
+                "flavorStr": "Pre-Kernel :",
+                "measurement": f"{nodeName}_ingress_dma_wait_measurement",
                 "profileIdxVar": profileIdxVar,
             })
 
@@ -166,9 +203,8 @@ def injectPrintCycleDiff(cls, executionBlock: ExecutionBlock, metaInfo: TilingMe
                 cls._printCycleDifference, {
                     "prefixStr": f"{nodeName}_prefix",
                     "suffixStr": f"{nodeName}_suffix",
-                    "flavorStr": "Kernel took ",
-                    "measurementsStart": f"{nodeName}_kernel_start_measurements",
-                    "measurementsEnd": f"{nodeName}_kernel_end_measurements",
+                    "flavorStr": "Kernel     :",
+                    "measurement": f"{nodeName}_kernel_measurement",
                     "profileIdxVar": profileIdxVar,
                 })
 
@@ -176,12 +212,23 @@ def injectPrintCycleDiff(cls, executionBlock: ExecutionBlock, metaInfo: TilingMe
             cls._printCycleDifference, {
                 "prefixStr": f"{nodeName}_prefix",
                 "suffixStr": f"{nodeName}_suffix",
-                "flavorStr": "Output DMA took ",
-                "measurementsStart": f"{nodeName}_egress_dma_wait_start_measurements",
-                "measurementsEnd": f"{nodeName}_egress_dma_wait_end_measurements",
+                "flavorStr": "Post-Kernel:",
+                "measurement": f"{nodeName}_egress_dma_wait_measurement",
                 "profileIdxVar": profileIdxVar,
             })
 
+        # Total Time: Input + Kernel + Output
+        # Overhead: (Input + Output) / Total
+        if metaInfo.kernelLevelTiling:
+            executionBlock.addRight(
+                cls._printCycleContribution, {
+                    "prefixStr": f"{nodeName}_prefix",
+                    "measurementInput": f"{nodeName}_ingress_dma_wait_measurement",
+                    "measurementKernel": f"{nodeName}_kernel_measurement",
+                    "measurementOutput": f"{nodeName}_egress_dma_wait_measurement",
+                    "profileIdxVar": profileIdxVar,
+                })
+
         executionBlock.addRight(cls._printLoopTeardown, {})
 
         return executionBlock