With the development of novel sequencing technologies, the way has been paved for cost efficient, high-throughput whole genome sequencing. In the year 2008 alone, about 250 genomes will have been sequenced. It is self-evident that the handling of this wealth of data requires efficient and accurate computational methods for sequence analysis. They are needed to tackle one of the most important problems in computational biology: the localisation of genes on DNA. In this thesis, I describe the development of state-of-the-art genomic signal detectors based on Support Vector Machines (SVM) that can be used in gene finding systems. The main contributions of this work are computationally efficient and accurate String Kernels, Large Scale Learning Methods, Methods to interprete SVM Classifiers.