This paper presents a framework for whole-body motion control of a free-standing humanoid robot, for kinesthetic teaching through pHRI (Physical Human Robot Interaction). The proposed method is based on a task-oriented control through task prioritization which enables execution of multiple robot tasks in parallel, exploiting system redundancy. The tasks with lower priority are handled without affecting execution of the primary ones. The proposed control algorithm provides compliant behavior of the robot end-effectors, in order to enable their physical guiding by demonstrator for skill acquisition, while preserving dynamic balance of a free standing humanoid, by keeping the zero-moment point (ZMP) inside the support area, in the presence of multiple constraints and contacts. The framework takes into account torque saturation, joints' mechanical limits and exploits system redundancy to maintain the desired realization of imposed tasks. In the proposed approach all imposed tasks and constraints, including the maintenance of dynamic balance, are handled without direct influence of a human, which allows a user to focus on the end-effector guidance. The proposed method has been tested by a simulation of manipulation task on a humanoid robot. Hip-ankle strategy, for the maintenance of dynamic balance, emerges as a consequence, without explicit programing.