Safety-critical autonomous systems require controllers that achieve performance objectives while guaranteeing constraint satisfaction. This paper presents a framework integrating Discrete Event System (DES) supervisory control with reinforcement learning (RL) through Control Barrier Function (CBF) principles. We establish a formal connection between DES state avoidance and CBF-based safety by defining discrete states as approximations of CBF level sets. The proposed DES-RL framework employs barrier-inspired reward shaping that encodes safety requirements into the learning process, enabling agents to discover optimal policies while maintaining forward invariance of safe sets. We develop a Q-learning algorithm with CBF-shaped rewards that learns safe policies without requiring explicit system dynamics models. The framework is validated through an autonomous vehicle lane-keeping application, demonstrating zero lane departure violations across 100 evaluation episodes. Comparative analysis with traditional CBF-QP controllers shows that DES-RL achieves equivalent safety under nominal to strong disturbance conditions.