This paper introduces a multi-layer software architecture, that allows smart design and implementation of complex cortical neural networks, and efficient parallel execution on multiprocessor machines. The developer implements cortical networks sticking to their natural fine grained formalism, without caring about its mapping on coarse grained parallel computers. The developer can use a set of graphical tools to easily develop and debug the cortical systems. Some experiments of new cortical model design on multiprocessor PCs are introduced, and some performance measurements are given. This software suite is operational and currently in use in our laboratory to control a robotic am through cortical neural networks running on multiprocessor PCs.