diff --git a/src/runtime/NEON/functions/NEPoolingLayer.cpp b/src/runtime/NEON/functions/NEPoolingLayer.cpp
index dd7a3a3..1570cde 100644
--- a/src/runtime/NEON/functions/NEPoolingLayer.cpp
+++ b/src/runtime/NEON/functions/NEPoolingLayer.cpp
@@ -23,7 +23,9 @@
  */
 #include "arm_compute/runtime/NEON/functions/NEPoolingLayer.h"
 
+#include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Validate.h"
+#include "arm_compute/runtime/Tensor.h"
 #include "src/runtime/cpu/operators/CpuPooling.h"
 
 namespace arm_compute
@@ -33,16 +35,15 @@
     ITensor                         *src{ nullptr };
     ITensor                         *dst{ nullptr };
     ITensor                         *indices{ nullptr };
-    std::shared_ptr<IMemoryManager>  memory_manager{ nullptr };
+    Tensor                           workspace{ nullptr };
     std::unique_ptr<cpu::CpuPooling> op{ nullptr };
 };
 
 NEPoolingLayer::~NEPoolingLayer() = default;
 
 NEPoolingLayer::NEPoolingLayer(std::shared_ptr<IMemoryManager> memory_manager)
-    : _impl(std::make_unique<Impl>())
+    : _memory_group(memory_manager), _impl(std::make_unique<Impl>())
 {
-    _impl->memory_manager = std::move(memory_manager);
 }
 
 void NEPoolingLayer::configure(ITensor *input, ITensor *output, const PoolingLayerInfo &pool_info, ITensor *indices)
@@ -50,8 +51,17 @@
     _impl->src     = input;
     _impl->dst     = output;
     _impl->indices = indices;
-    _impl->op      = std::make_unique<cpu::CpuPooling>(_impl->memory_manager);
+    _impl->op      = std::make_unique<cpu::CpuPooling>();
     _impl->op->configure(input->info(), output->info(), pool_info, (indices) ? indices->info() : nullptr);
+
+    // Allocate workspace based on kernel's memory requirements
+    const experimental::MemoryRequirements mem_req = _impl->op->workspace();
+    if(!mem_req.empty())
+    {
+        _impl->workspace.allocator()->init(TensorInfo(TensorShape{ (mem_req[0].size + mem_req[0].alignment) }, 1, DataType::S8), mem_req[0].alignment);
+        _memory_group.manage(&_impl->workspace);
+        _impl->workspace.allocator()->allocate();
+    }
 }
 
 Status NEPoolingLayer::validate(const ITensorInfo *input, const ITensorInfo *output, const PoolingLayerInfo &pool_info, const ITensorInfo *indices)
@@ -65,6 +75,7 @@
     pack.add_tensor(TensorType::ACL_SRC, _impl->src);
     pack.add_tensor(TensorType::ACL_DST_0, _impl->dst);
     pack.add_tensor(TensorType::ACL_DST_1, _impl->indices);
+    pack.add_tensor(TensorType::ACL_INT_0, &_impl->workspace);
     _impl->op->run(pack);
 }
 } // namespace arm_compute
diff --git a/src/runtime/cpu/operators/CpuPooling.cpp b/src/runtime/cpu/operators/CpuPooling.cpp
index 0b9b38d..3a6ac24 100644
--- a/src/runtime/cpu/operators/CpuPooling.cpp
+++ b/src/runtime/cpu/operators/CpuPooling.cpp
@@ -27,41 +27,59 @@
 #include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/runtime/NEON/NEScheduler.h"
 #include "src/core/NEON/kernels/NEFillBorderKernel.h"
+#include "src/core/cpu/kernels/CpuPoolingAssemblyWrapperKernel.h"
 #include "src/core/cpu/kernels/CpuPoolingKernel.h"
 
 namespace arm_compute
 {
 namespace cpu
 {
-CpuPooling::CpuPooling(std::shared_ptr<IMemoryManager> memory_manager)
-    : _memory_manager(std::move(memory_manager)), _pooling_layer_kernel(), _border_handler(), _asm_glue(), _is_global_pooling_layer(false), _data_layout(DataLayout::NCHW)
+CpuPooling::CpuPooling()
+    : _pooling_layer_kernel(),
+      _border_handler(),
+      _asm_glue(),
+      _is_global_pooling_layer(false),
+      _data_layout(DataLayout::NCHW),
+      _mem_req()
 {
 }
 
 CpuPooling::~CpuPooling() = default;
 
-void CpuPooling::configure(ITensorInfo *input, ITensorInfo *output, const PoolingLayerInfo &pool_info, ITensorInfo *indices)
+void CpuPooling::configure(ITensorInfo *src, ITensorInfo *dst, const PoolingLayerInfo &pool_info, ITensorInfo *indices)
 {
     // Check if we can run assembly kernels. Currently, indices are not supported by those kernels
-    const bool run_optimised = bool(CpuPoolingAssemblyDispatch::validate(input, output, pool_info)) && (indices == nullptr);
+    const bool run_optimised = bool(kernels::CpuPoolingAssemblyWrapperKernel::validate(src, dst, pool_info)) && (indices == nullptr);
+
+    // Get data layout
+    _data_layout = pool_info.data_layout == DataLayout::UNKNOWN ? src->data_layout() : pool_info.data_layout;
+
+    // Check if we have Global Pooling Layer
+    const unsigned int idx_width  = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::WIDTH);
+    const unsigned int idx_height = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::HEIGHT);
+    _is_global_pooling_layer      = (src->dimension(idx_width) == pool_info.pool_size.width) && (src->dimension(idx_height) == pool_info.pool_size.height);
 
     if(run_optimised)
     {
-        _asm_glue = std::make_unique<CpuPoolingAssemblyDispatch>(_memory_manager);
-        _asm_glue->configure(input, output, pool_info);
-        ARM_COMPUTE_ERROR_ON(!_asm_glue->is_configured());
+        const CPUInfo     &ci          = NEScheduler::get().cpu_info();
+        const unsigned int num_threads = NEScheduler::get().num_threads();
+
+        auto pooling_wrapper = std::make_unique<kernels::CpuPoolingAssemblyWrapperKernel>();
+        ARM_COMPUTE_ERROR_ON(pooling_wrapper == nullptr);
+        pooling_wrapper->configure(src, dst, pool_info, ci);
+
+        // Get kernel's memory requirements
+        constexpr size_t alignment      = 4096;
+        const size_t     workspace_size = pooling_wrapper->get_working_size(num_threads);
+        _mem_req.push_back({ TensorType::ACL_INT_0, workspace_size, alignment });
+
+        _asm_glue = std::move(pooling_wrapper);
     }
     else
     {
-        // Check if we have Global Pooling Layer
-        _is_global_pooling_layer = (input->dimension(0) == pool_info.pool_size.width) && (input->dimension(1) == pool_info.pool_size.height);
-
-        // Get data layout
-        _data_layout = pool_info.data_layout == DataLayout::UNKNOWN ? input->data_layout() : pool_info.data_layout;
-
         // Configure pooling kernel
         auto k = std::make_unique<kernels::CpuPoolingKernel>();
-        k->configure(input, output, pool_info, indices);
+        k->configure(src, dst, pool_info, indices);
         _pooling_layer_kernel = std::move(k);
 
         switch(_data_layout)
@@ -71,12 +89,12 @@
                 // Configure border depending on operation required (quantize border in case of asymmetric data_type)
                 BorderMode border_mode = (!indices && pool_info.pool_type == PoolingType::MAX) ? BorderMode::REPLICATE : BorderMode::CONSTANT;
                 PixelValue zero_value((indices) ? std::numeric_limits<int>::min() : 0.f);
-                if(is_data_type_quantized_asymmetric(input->data_type()) && !pool_info.exclude_padding)
+                if(is_data_type_quantized_asymmetric(src->data_type()) && !pool_info.exclude_padding)
                 {
-                    zero_value = PixelValue(0, input->data_type(), input->quantization_info());
+                    zero_value = PixelValue(0, src->data_type(), src->quantization_info());
                 }
                 auto b = std::make_unique<NEFillBorderKernel>();
-                b->configure(input, _pooling_layer_kernel->border_size(), border_mode, zero_value);
+                b->configure(src, _pooling_layer_kernel->border_size(), border_mode, zero_value);
                 _border_handler = std::move(b);
                 break;
             }
@@ -88,23 +106,26 @@
     }
 }
 
-Status CpuPooling::validate(const ITensorInfo *input, const ITensorInfo *output, const PoolingLayerInfo &pool_info, const ITensorInfo *indices)
+Status CpuPooling::validate(const ITensorInfo *src, const ITensorInfo *dst, const PoolingLayerInfo &pool_info, const ITensorInfo *indices)
 {
-    const bool run_optimised = bool(CpuPoolingAssemblyDispatch::validate(input, output, pool_info)) && (indices == nullptr);
+    const bool run_optimised = bool(kernels::CpuPoolingAssemblyWrapperKernel::validate(src, dst, pool_info)) && (indices == nullptr);
 
     if(run_optimised)
     {
         return Status{};
     }
 
-    return kernels::CpuPoolingKernel::validate(input, output, pool_info, indices);
+    return kernels::CpuPoolingKernel::validate(src, dst, pool_info, indices);
 }
 
 void CpuPooling::run(ITensorPack &tensors)
 {
-    if(_asm_glue && _asm_glue->is_configured())
+    ARM_COMPUTE_ERROR_ON_MSG(tensors.empty(), "No tensors provided");
+
+    if(_asm_glue)
     {
-        _asm_glue->run(tensors);
+        const auto hints = (_is_global_pooling_layer) ? Window::DimX : Window::DimY;
+        NEScheduler::get().schedule_op(_asm_glue.get(), hints, _asm_glue->window(), tensors);
     }
     else
     {
@@ -126,5 +147,10 @@
         }
     }
 }
+
+experimental::MemoryRequirements CpuPooling::workspace() const
+{
+    return _mem_req;
+}
 } // namespace cpu
 } // namespace arm_compute
diff --git a/src/runtime/cpu/operators/CpuPooling.h b/src/runtime/cpu/operators/CpuPooling.h
index b1647ea..bc30adf 100644
--- a/src/runtime/cpu/operators/CpuPooling.h
+++ b/src/runtime/cpu/operators/CpuPooling.h
@@ -26,8 +26,7 @@
 
 #include "src/runtime/cpu/ICpuOperator.h"
 
-#include "arm_compute/runtime/IMemoryManager.h"
-#include "src/runtime/cpu/operators/CpuPoolingAssemblyDispatch.h"
+#include "arm_compute/core/experimental/Types.h"
 
 #include <memory>
 
@@ -38,19 +37,17 @@
 
 namespace cpu
 {
-// Forward Declarations
-class CpuPoolingAssemblyDispatch;
 /** Basic function to simulate a pooling layer with the specified pooling operation. This function calls the following kernels:
  *
  * -# @ref NEFillBorderKernel (executed if padding size is different from zero)
  * -# @ref kernels::CpuPoolingKernel
- * -# @ref CpuPoolingAssemblyDispatch
+ * -# @ref kernels::CpuPoolingAssemblyWrapperKernel
  */
 class CpuPooling : public ICpuOperator
 {
 public:
     /** Constructor */
-    CpuPooling(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    CpuPooling();
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     CpuPooling(const CpuPooling &) = delete;
     /** Prevent instances of this class from being copied (As this class contains pointers) */
@@ -86,16 +83,16 @@
 
     // Inherited methods overridden:
     void run(ITensorPack &tensors) override;
+    experimental::MemoryRequirements workspace() const override;
 
 private:
-    std::shared_ptr<IMemoryManager> _memory_manager;
+    std::unique_ptr<INEKernel> _pooling_layer_kernel;
+    std::unique_ptr<INEKernel> _border_handler;
+    std::unique_ptr<INEKernel> _asm_glue;
 
-    std::unique_ptr<INEKernel>                  _pooling_layer_kernel;
-    std::unique_ptr<INEKernel>                  _border_handler;
-    std::unique_ptr<CpuPoolingAssemblyDispatch> _asm_glue;
-
-    bool       _is_global_pooling_layer;
-    DataLayout _data_layout;
+    bool                             _is_global_pooling_layer;
+    DataLayout                       _data_layout;
+    experimental::MemoryRequirements _mem_req;
 };
 } // namespace cpu
 } // namespace arm_compute
diff --git a/src/runtime/cpu/operators/CpuPoolingAssemblyDispatch.cpp b/src/runtime/cpu/operators/CpuPoolingAssemblyDispatch.cpp
deleted file mode 100644
index e4526c5..0000000
--- a/src/runtime/cpu/operators/CpuPoolingAssemblyDispatch.cpp
+++ /dev/null
@@ -1,98 +0,0 @@
-/*
- * Copyright (c) 2021 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#include "src/runtime/cpu/operators/CpuPoolingAssemblyDispatch.h"
-
-#include "arm_compute/core/ITensor.h"
-#include "arm_compute/runtime/NEON/NEScheduler.h"
-#include "src/core/CPP/Validate.h"
-#include "src/core/cpu/kernels/CpuPoolingAssemblyWrapperKernel.h"
-
-namespace arm_compute
-{
-namespace cpu
-{
-CpuPoolingAssemblyDispatch::CpuPoolingAssemblyDispatch(std::shared_ptr<IMemoryManager> memory_manager)
-    : _memory_group(std::move(memory_manager)),
-      _workspace(),
-      _is_global_pooling_layer(false)
-{
-}
-
-CpuPoolingAssemblyDispatch::~CpuPoolingAssemblyDispatch() = default;
-
-void CpuPoolingAssemblyDispatch::configure(const ITensorInfo *src, ITensorInfo *dst, const PoolingLayerInfo &info)
-{
-    const CPUInfo     &ci          = NEScheduler::get().cpu_info();
-    const unsigned int num_threads = NEScheduler::get().num_threads();
-
-    // If we don't support a combination of data types, silently return: it is the caller's responsibility to check if configure() was successful via is_configured()
-    if(!CpuPoolingAssemblyDispatch::validate(src, dst, info))
-    {
-        return;
-    }
-
-    auto pooling_wrapper = std::make_unique<kernels::CpuPoolingAssemblyWrapperKernel>();
-    ARM_COMPUTE_ERROR_ON(pooling_wrapper == nullptr);
-    pooling_wrapper->configure(src, dst, info, ci);
-
-    // Check if we have Global Pooling Layer
-    _is_global_pooling_layer = (src->dimension(2) == info.pool_size.width) && (src->dimension(1) == info.pool_size.height);
-
-    // Allocate workspace based on kernel's memory requirements
-    constexpr size_t alignment      = 4096;
-    const size_t     workspace_size = pooling_wrapper->get_working_size(num_threads);
-    _workspace.allocator()->init(TensorInfo(TensorShape{ (workspace_size + alignment) }, 1, DataType::S8), alignment);
-    _memory_group.manage(&_workspace);
-    _workspace.allocator()->allocate();
-
-    _kernel = std::move(pooling_wrapper);
-}
-
-Status CpuPoolingAssemblyDispatch::validate(const ITensorInfo *src, const ITensorInfo *dst, const PoolingLayerInfo &info)
-{
-    return kernels::CpuPoolingAssemblyWrapperKernel::validate(src, dst, info);
-}
-
-bool CpuPoolingAssemblyDispatch::is_configured() const
-{
-    return _kernel != nullptr;
-}
-
-void CpuPoolingAssemblyDispatch::run(ITensorPack &tensors)
-{
-    ARM_COMPUTE_ERROR_ON_MSG(tensors.empty(), "No srcs provided");
-
-    tensors.add_tensor(TensorType::ACL_DST_1, &_workspace);
-
-    if(_is_global_pooling_layer)
-    {
-        NEScheduler::get().schedule_op(_kernel.get(), Window::DimX, _kernel->window(), tensors);
-    }
-    else
-    {
-        NEScheduler::get().schedule_op(_kernel.get(), Window::DimY, _kernel->window(), tensors);
-    }
-}
-} // namespace cpu
-} // namespace arm_compute
diff --git a/src/runtime/cpu/operators/CpuPoolingAssemblyDispatch.h b/src/runtime/cpu/operators/CpuPoolingAssemblyDispatch.h
deleted file mode 100644
index 353bbe1..0000000
--- a/src/runtime/cpu/operators/CpuPoolingAssemblyDispatch.h
+++ /dev/null
@@ -1,89 +0,0 @@
-/*
- * Copyright (c) 2021 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CPU_POOLING_ASSEMBLY_DISPATCH_H
-#define ARM_COMPUTE_CPU_POOLING_ASSEMBLY_DISPATCH_H
-
-#include "arm_compute/core/Types.h"
-#include "arm_compute/runtime/IMemoryManager.h"
-#include "arm_compute/runtime/MemoryGroup.h"
-#include "arm_compute/runtime/Tensor.h"
-#include "src/runtime/cpu/ICpuOperator.h"
-
-namespace arm_compute
-{
-namespace cpu
-{
-class ITensor;
-
-/** Basic function to run pooling assembly kernels */
-class CpuPoolingAssemblyDispatch : public ICpuOperator
-{
-public:
-    /** Constructor */
-    CpuPoolingAssemblyDispatch(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
-    /** Prevent instances of this class from being copied */
-    CpuPoolingAssemblyDispatch(const CpuPoolingAssemblyDispatch &) = delete;
-    /** Default move constructor */
-    CpuPoolingAssemblyDispatch(CpuPoolingAssemblyDispatch &&) = default;
-    /** Prevent instances of this class from being copied */
-    CpuPoolingAssemblyDispatch &operator=(const CpuPoolingAssemblyDispatch &) = delete;
-    /** Default move assignment operator */
-    CpuPoolingAssemblyDispatch &operator=(CpuPoolingAssemblyDispatch &&) = default;
-    /** Destructor */
-    ~CpuPoolingAssemblyDispatch();
-
-    /** If supported create an assembly routine, else fallback to Compute Library function.
-     *
-     * @param[in]  src  Source tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32.
-     * @param[out] dst  Destination tensor info to store the result of pooling. Data types supported: same as @p src.
-     * @param[in]  info Pooling meta-data
-     */
-    void configure(const ITensorInfo *src, ITensorInfo *dst, const PoolingLayerInfo &info);
-
-    /** Indicates whether or not this function can be used to process the given parameters.
-     *
-     * @param[in] src  Source tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32.
-     * @param[in] dst  Destination tensor to store the result of pooling. Data types supported: same as @p src.
-     * @param[in] info Pooling meta-data
-     *
-     * @return a status.
-     */
-    static Status validate(const ITensorInfo *src, const ITensorInfo *dst, const PoolingLayerInfo &info);
-    /** Was the function successfully configured ?
-     *
-     * @return True if the function is configured and ready to run
-     */
-    bool is_configured() const;
-    // Run method overriden
-    void run(ITensorPack &tensors) override;
-
-private:
-    arm_compute::MemoryGroup _memory_group;
-
-    arm_compute::Tensor _workspace;
-    bool                _is_global_pooling_layer;
-};
-} // namespace cpu
-} // namespace arm_compute
-#endif /* ARM_COMPUTE_CPU_POOLING_ASSEMBLY_DISPATCH_H */