Developed and integrated a new GEMM post-processing step called 'splitk' within the intel/sycl-tla repository, enabling the output tensor to be divided into two distinct parts, 'nope' and 'rope', based on configurable dimensions. This feature allows for specialized downstream handling and opens opportunities for targeted optimizations in high-performance tensor workflows. The implementation leveraged C++, SYCL, and CMake, focusing on modularity and alignment with the project’s performance roadmap. By tracking the work under a dedicated issue, the developer ensured clear documentation and roadmap alignment, laying the groundwork for future enhancements in GPU programming and high-performance computing environments.