Developing manipulators for kinesthetic haptic interfaces is challenging due to a large number of design parameters. We propose a novel optimization-driven design approach taking into account the properties of the entire workspace of the human arm instead of a specific task. To achieve this, models of both the human arm and the haptic manipulator are derived and deployed in a suitable objective function, which simultaneously considers poses, velocities, accelerations, as well as displayed forces and torques. A detailed analysis and experiments with real-world motion tracking data show that the proposed method is capable of finding meaningful design parameters to enable good haptic transparency.