This thesis proposes a new approach to improve multiclass classifica-tion performance by employing Stacked Generalization structure and One-Against-One decomposition strategy. The proposed approach en-codes the outputs of all pairwise classifiers by implicitly embedding two-class discriminative information in a probabilistic manner. The encoded outputs, called Meta Probability Codes (MPCs), are inter-preted as the projections of the original features. It is shown that the MPC, compared to the original features, has more appropriate fea-tures for clustering. Based on MPC features, we introduce a cluster-based multiclass classification algorithm, called MPC-Clustering. The MPC-Clustering algorithm uses the proposed approach to project an original feature space to MPC, and then it employs a clustering scheme to cluster MPCs. Subsequently, it trains individual multi-class classifiers on the produced clusters to complete the procedure of multiclass classifier induction.The performance of our proposed algorithm is evaluated by applying it on20different datasets from the UCI machine learning database repository. It is shown that our algorithm improves the classification rate by almost2.4%on average. Moreover, the performance of the projected features is also evaluated without applying a clustering step. That is, a known multiclass classifier is trained directly on the pro-jected samples. It is shown that the classification accuracy of SVM and k-NN trained on the projected features improved by0.99%and3.62%, respectively.In this thesis, we also study the performance of the MPC features on two real world applications, face and facial expression recognition via proposing an MPC-based framework, in which any feature extractor and classifier can be incorporated in the proposed framework using the meta-feature generation mechanism. In the experimental studies, we use some of the state-of-the-art and promising multiclass classi-fiers and information representation techniques. The results of the extensive experiments conducted on three facial expression datasets; Cohn-Kanade, JAFFE and TFEID, and two face recognition datasets; FERET and CAS-PEAL-R1, show that the MPC features promote the performance of face and facial expression recognition inherently.