This paper addresses the problem of load-shaving in Smart Homes in view of improving energy efficiency in Smart Grids. An architecture of home power management systems, allowing the separation of domestic power load control from grid dynamics, is introduced. In this framework, the operation of appliances is modeled as a finite state machine which enables the implementation of online scheduling borrowed from the techniques developed in real-time computing systems. A scheduling algorithm is developed for peak-load shaving and the simulation results confirm the effectiveness and the efficiency of the proposed approach.