Significant efforts have been put into optimization of GALS systems by considering locally synchronous modules as individual and independent islands of circuits. While existing approaches do improve some of the system characteristics, due to their limited scope, their achieved improvement is often limited too. This paper proposes an optimization approach in which a GALS system is optimized as a whole. The approach allows combinational or sequential sub-modules to move from one synchronous module to another while preserving the GALS system functionality. Experimental results show that if the movements are done properly, the proposed approach provides better results than the existing methods. The application of the proposed approach on benchmark circuits demonstrates 15.8% latency reduction with 3.3% area overhead in average.