In this study, we propose a novel block coordinate DCA based method for tackling the large-scale kernel SVM. The proposed method employs a unified scheme that is capable of handling almost all common losses in SVM. Owing to the block coordinate approach, the learning algorithm can deal with the high dimensionality of the optimization variables by updating one block of coordinates at a time to sufficiently decrease the objective value while keeping the other blocks fixed. As a result, calculation time can be significantly reduced while preventing the computer’s memory from overflowing, which is crucial for computing in the context of big data. Numerical experiments are conducted intensively, which shows the algorithm’s merits in both accuracy and computational cost.