In this work, a novel adaptive approach to co-design of embedded systems is presented. The approach is based on developmental genetic programming. Unlike most of existing algorithms, presented methodology involves evolving co-synthesis process, not the system architecture directly. Genotype is a tree which nodes include system construction options. The system can adapt to the environment by increasing chromosomes which give better results in each situations. Half of the next populations is created using genetic operators (crossover, mutation, reproduction). Second half is obtained by generating additional solutions but with different probability of the options.