To enhance Human interaction with machines, the research interest is increasing in the field of Brain Computer Interaction which allows people to communicate with external systems just by their mental activity. Until now the applicability of Brain Computer Interface has been strongly restricted by low bit-transfer rates, slow response times and long training sessions for the subject. There is a need to improve both classification performance and reduce the need of subject training. This paper discusses the effectiveness and accuracy of the proposed novel approach for Classifying three Non-Movement-Mental-Tasks namely- Math task, Counting Task and Idle Mental Task through a Wavelet decomposition of EEG Signals and then classifying the selected features of Power Spectral and Power spectral Difference using a new classifier system which incorporates Modern as well as Classical Artificial Intelligence. In the Classical AI we have used deduction based classification, for which we introduced a new concept of Voting among Segmental-Components of any EEG trial while the modern AI was based on Support Vector Machine (SVM). The key motivation of this paper has been the improvement in the following two situations – 1) Finding out the perfect feature for classification of Segmental Samples and 2) Increasing the accuracy with respect to classification of actual Samples or Trials instead of Segmental Samples. According to the experimental results we have confirmed the feasibility of the proposed novel approach by comparing the results with the previous research results.