High-speed vision sensing becomes a driving factor in developing new methods for robotic manipulation. In this paper we present two such methods in order to realize high-performance manipulation. First, we present a dynamic compensation approach which aims to achieve simultaneously fast and accurate positioning under various (from system to external environment) uncertainties. Second, a high-speed motion strategy for manipulating flexible objects is introduced to address the issue of deformation uncertainties. Both methods rely on high-speed visual feedback and are model independent, which we believe is essential to ensure good flexibility in a wide range of applications. The high-speed visual feedback tracks the relative error between the working tool and the target in image coordinates, which implies that there is no need for accurate calibrations of the vision system. Tasks for validating these methods were implemented and experimental results were provided to illustrate the effectiveness of the proposed methods.