Towards Automated Kernel Generation in the Era of LLMs