Pattern analysis is an integral part of many scientific disciplines such as machine learning, data mining, statistics and bioinformatics. In this thesis, we have observed how to apply kernel methods in different learning algorithms and how to find patterns between data. We have seen that such approach is modular, any kernel method can be combined with any algorithm and vice versa. The modularity and the fact that the approach was able to bridge the gaps that existed between the different disciplines in which pattern analysis is used, have provided us with a unique tool for working on data of all types, whether they are vectors, sets, strings, or even a complex form of data. In the first chapter we have described the problem and we have outlined the mathematical definitions we have used in the following text to make it easier to understand. In the second chapter we introduced the term kernel method in a concrete example. We’ve seen how to find linear relations in some finite data set, and we’ve seen the benefits of using them. Then we’ve made the classification of the kernel methods according to the type of data they are intended for. In the third chapter, we have described efficient learning algorithms for novelty detection, classification and regression based on the support vectors. In the fourth chapter, we implemented two classification algorithms with two different kernels and applied them to the problem of text categorization. From the obtained results, we concluded that the robustness of the algorithm greatly influences the success of the classification. A more robust algorithm was far more successful. We have seen that the choice of the kernels and the selection of their parameters also influenced the success of algorithms.