By PCA did you mean Principal component analysis?
If so, PCA does not extract features, it evaluates their correlation and indicates the more useful ones. PCA is employed for feature selection, no feature extraction. It should be done according the expertise, the case of study, and the features of interest.
PCA will indicate which features would be more useful as classification criteria.
I recommend you to check documentation of other toolboxes related to classification and feature extraction (machine learning and statitics toolbox) and PM toolbox, instead of only information regarding signal processing toolbox.