In the embedded system design, memory is one of the most restricted resources. Code compression has been proposed as a solution to reduce the code size of applications for embedded systems. Data compression techniques are used to compress programs to reduce memory size. Most previous work compresses all instructions found in an executable, without taking into account the program execution profile. In this paper, a profile-driven code compression design methodology is proposed. Program profiling information can be used to help code compression to selectively compress non-critical instructions, such that the system performance degradation due to the decompression penalty is reduced.