Support vector classification plays a vital role in the discrimination and classification of observations/objects into the corresponding groups in the domain of supervised learning techniques. In this context kernel functions have a significant impact on computational aspects. Many kernel functions are providing reliable results when cleaned/processed data is in the experimental study. Hence, it is much more essential in selecting the suitable kernel function for getting a more accurate classification. An attempt has been made to explore the most widely used kernel functions and studied their robustness. This paper explores the various kernel functions and compares their efficiency by computing the classification rate on real datasets. Further, the simulation study has been carried out along with the various contaminations (location/scale/location and scale). The real and simulation environment shows that the radial kernel approach outperforms the others in terms of prediction accuracy.