Task scheduling is the primary multitasking activity controlled by the real-time executive. As hardware/software co-design of embedded systems has been enabled by advances in computer technologies, reprogrammable hardware can be used to implement a co-processor to perform most of the kernel functions, including task scheduling. In this kind of system design, more complex scheduling approaches can be applied. In this paper, a complex scheduling approach which takes advantages of evolutionary computation (i.e., neurocomputing and genetic search and optimization) into account is presented. First, we present a model based on the Hopfield-Tank neural network. Then, we introduce modifications of the method based on the network model to improve the quality of the solutions. Finally, we propose a mixed approach of this evolutionary computation method and an extension of the Earliest Deadline First approach for scheduling both types of periodic and aperiodic tasks. We also discuss simulation results that demonstrate performance that could be obtained by using this approach.