The proposed work is done in order to develop an optimised Brain-Computer Interface (BCI) system (speller) for people with severe motor impairments using SSVEP (Steady-State Visual Evoked Potentials). To make the system fast yet error-free, the optimisation of speller is divided into three domains: one is the design of smart encoding method for the selection of appeared characters on interface, second one is the optimal frequency choice and the last one is design of optimal feature classification algorithm. Three classification methods: threshold method, Artificial Neural Network (ANN) and Support Vector Machine (SVM) are evaluated. An optimal user window is also carefully selected after many trails in order to maintain a decent communication rate. The optimised BCI system provides an average accuracy of 96% with character per minute (CPM) of 13 � 2. Speller performs almost similar with new users too because inter-subject variability is tackle by SVM classifier.