Self-Evolved Reward Learning for LLMs