Configurable system-on-chip often uses a soft-core programmable processor, running a real-time operating system (RTOS). However, the CPU overheads imposed by the RTOS increase rapidly with the number of tasks in the system. In This work, we propose instruction set customization as the means to contain RTOS-imposed CPU overheads. We present the design of custom instructions for frequently used RTOS primitives and present our area-time results for a variety of configurations. We show that complex embedded systems stand to benefit significantly from our approach. Individual RTOS routines show performance improvements in the range of 50-90%. Frequently used RTOS primitives showed an improvement of 10-35% and the Dhrystone mark of the system improved by as much as 13%.