Support Vector Machines (SVMs) are among the most widely used machine learning techniques, recognized for their strong ability to handle high-dimensional data and construct robust decision boundaries. Nevertheless, applying them to multiclass or nonlinear classification tasks continues to pose significant challenges, particularly in medical domains where data are often complex and highly variable. Building on this context, the present study introduces an integrative approach that incorporates the Gaussian pyramid as a preprocessing stage prior to SVM training. The aim of this integration is to reduce noise, eliminate redundant features, and preserve the essential discriminative characteristics that separate different classes. Furthermore, this method enhances the stability of decision boundaries and reduces the number of required support vectors, which in turn accelerates the training process and improves efficiency. The proposed approach was evaluated on three benchmark medical datasets: sleep stage classification, cardiac arrhythmia detection, and heart sound recognition. Experimental findings revealed that embedding the Gaussian pyramid into the workflow reduced training time by 30% to 50% compared with the conventional SVM model, while maintaining high levels of predictive accuracy. The best performance was recorded with a 5×5 kernel applied to the sleep stage dataset, achieving an accuracy of 0.96 and an area under the curve of 0.98. However, increasing the kernel size to 7×7 or 9×9 led to a decline in performance due to the loss of fine-grained features. This highlights the sensitivity of the method to kernel size selection and underscores the importance of carefully balancing smoothing and feature preservation. Keywords: Multi-class SVM, Gaussian pyramid, Nonlinear SVM, Large datasets.