Support Vector Machines (SVM) is a popular kernel-based method for data classifica- 2 tion that have demonstrated high efficiency across a wide range of practical applications. 3 However, SVM suffers from several limitations, including the potential failure of the opti- 4 mization process,especially in high-dimensional spaces, the inherently high computational 5 cost, the lack of a systematic approach to multiclass classification, difficulties in handling 6 imbalanced classes, and the prohibitive cost of real-time or dynamic classification. This 7 paper proposes an alternative method, referred to as Kernel-based Optimal Subspaces 8 (KOS). The method achieves performance comparable to SVM while addressing the afore- 9 mentioned weaknesses. It is based on computing a minimum distance to optimal feature 10 subspaces of the mapped data. No optimization process is required, which makes the 11 method robust, fast, and easy to implement. The optimal subspaces are constructed inde- 12 pendently, enabling high parallelizability and making the approach well-suited for dynamic 13 classification and real-time applications. Furthermore, the issue of imbalanced classes is 14 naturally handled by subdividing large classes into smaller sub-classes, thereby creating 15 appropriately sized sub-subspaces within the feature space.