Humanoid robots are the most illustrative representatives of complex cyber-physical systems. They are high-dimensional and highly coupled non-linear systems. Moreover, walking humanoid robots are underactuated with friction-limited unilateral contacts with the ground. These facts justify the need for hard real-time control of a robot's motion. Even the smallest delay in calculating the reference motion or setting the output control variable can lead to a catastrophic outcome. For this reason, most of the existing robotic control systems are real-time and custom made, intended for a specific robot (or family of robots). There is no platform offering a general tool to design the model of versatile configurations of robots and to use it in the model-based control algorithms in hard real-time. In this paper, a software tool for modelling of an articulated system consisting of several open branched kinematic chains for a hard real-time OS is proposed. The model of the robot can possess an arbitrary number of link. The software is developed on top of the QNX Neutrino real-time operating system, thus providing the precise execution of time-critical events. In the experiment, the advantages of using the general model by modelling one real robotic configuration with multiple branched kinematic chains will be shown. The processing performance during the on-line motion planning will be tested on a real-time platform running on ARM Cortex-A8 processor. The results will be compared with similar software running in a non-real-time simulation environment.