Reducing power consumption is an important design objective in energy-constrained embedded and mobile systems. Such systems must be designed to meet functional and often timing requirements facing the challenge of energy restriction. This paper presents a new approach for power management based on ZETA, a mathematical framework that utilizes computational and energy models to provide proofs for optimal clock rate control. Special runtime models are developed to provide optimal values for the CPU clock rate under real-time requirements. The method is suitable for embedded systems which utilize a single task real-time computational model and linear energy models. A special tool called CASTLE is developed to support this approach. CASTLE helps us integrate power management into the operating system or to automatically insert power management into the application code. The tool utilizes as input application-specific data which is conveyed via comments in the source file.