Application-specific customization of the instruction set helps embedded processors achieve significant performance and power efficiency. In this paper, we explore customization in the context of multi-tasking real-time embedded systems. We propose efficient algorithms to select the optimal set of custom instructions for a task set under two popular real-time scheduling policies. Our algorithms minimize the processor utilization through customization while satisfying the task deadlines and the constraint on silicon area. Experimental evaluation with various task sets shows that appropriate customization can achieve significant reduction in the processor utilization and the energy consumption.