Computed torque is an effective method for control of robotic manipulators. In this paper, a novel approach is proposed for computed torque control method, and also, it is integrated with online least squares support vector regression algorithm. The motive is to improve control performance and to build an adaptive architecture where no prior information on system model is required; the model is obtained continuously using a machine learning technique. There are two main contributions of the proposed work: the first one is the formulation of the computed torque control law using the nonlinear autoregressive moving average-L2 control methodology and the second contribution is the implementation of online least squares support vector regression method in estimating the dynamical model of the system and calculating the control law. The performance of the proposed method has been evaluated by simulations on a two-link robotic arm. The robustness of the method has also been evaluated for cases with disturbance and model uncertainty. Comparison with an existing computed torque control approach in the literature demonstrates the superiority of the proposed approach. Thus, it can be concluded that the method can be effectively used for first-order and second-order nonlinear systems without inherent instability.