HPC clusters are characterized by powerful hardware and fine-tuned software environments to extract as much performance as possible from the system. Surprisingly, many HPC systems typically use the default kernel provided by its Linux distribution of choice, which is a generic binary built to maximize stability and compatibility and lacks the fine-tuning for performance characteristic of such systems. We explored the impact of different optimization techniques that can be carried out on the Linux kernel without changing its code. We tested compiler optimizations with GCC and LLVM, using CPU instruction sets specific to the target hardware, profile guided optimizations (PGO) or link time optimizations (LTO). The result is that such optimizations on kernel space have limited effects and depend on the type of workload, rendering them difficult to apply on HPC systems with a diverse ecosystem. The main goal is to generate (more) optimized kernel images with the same exact functionality of the distribution kernel used in your system. We specifically target kernels of RPM-based Linux distributions and show how to generate drop-in replacements with the aforementioned optimizations.