A new real-time control system will be implemented within the Keck II adaptive optics system to support the new near-infrared pyramid wavefront sensor. The new real-time computer has to interface with an existing, very productive adaptive optics system. We discuss our solution to install it in an operational environment without impacting science. This solution is based on an independent SCExAO-based pyramid wavefront sensor realtime processor solution using the hardware interfaces provided by the existing Keck II real-time controller. We introduce the new pyramid real-time controller system design, its expected performance, and the modification of the operational real-time controller to support the pyramid system including interfacing with the existing deformable and tip-tilt mirrors. We describe the integration of the Saphira detector-based camera and the Boston Micromachines kilo-DM in this new architecture. We explain the software architecture and philosophy, the shared memory concept and how the real-time computer uses the power of GPUs for adaptive optics control. We discuss the strengths and weaknesses of this architecture and how it can benefit other projects. The motion control of the devices deployed on the Keck II adaptive optics bench to support the alignment of the light on the sensors is also described. The interfaces, developed to deal with the rest of the Keck telescope systems in the observatory distributed system, are reviewed. Based on this experience, we present which design ideas could have helped us integrate the new system with the previous one and the resultant performance gains.