Soft robots are becoming increasingly popular because they are adaptable and safe. However, soft robot behaviors are extremely complex, so they can be hard to manually design. To resolve this, researchers have investigated automatic design processes for soft robots. Previous work has proposed evolution techniques targeted to single tasks. However, robots must perform multiple tasks to execute more complicated processes. We propose Combined-CPPN-NEAT. It is a method for designing robots that can perform multiple tasks by combining networks of robots that were first evolved to perform individual tasks. To investigate the method's performance, we ran two experiments. The first evolved morphologies with respect to predefined target functions, and the second comprised two locomotion tasks. Our results showed that the morphological and functional information of robots evolved to perform individual tasks can be extracted and combined using Combined-CPPN-NEAT. This method performed better than the existing multiple task evolution method.