In this paper, we present an approach for model-based movement and posture classification in working environments. The approach presented here is designed for long-term in-situ observations of and by workers in their workplaces. The proposed model is adaptable to different input data, eg, skeleton data from either an Inertial Measurement Unit (IMU) or a skeleton derived from an optical sensor such as Kinect. We present a preliminary design of the model and suggest algorithms suitable for real-time usage of the model in an IMU-based motion capture suite. In an experiment we measured the weight on the knee while performing different kneeing postures to show the dependence of posture angles on the knee load.