The accurate land cover mapping of the Earth's surface using Earth observation data is one of the most studied, but yet the most challenging tasks of remote sensing field, particularly when it comes to urban areas. The large spectral variability of man-made structures, as well as the mixed pixel phenomenon, imposes the use of computational demanding techniques, which are not always effective for real case applications. Support vector machines (SVMs) are supervised learning models with associated learning algorithms, which are mainly used for classification and regression analysis. Specifically, a support vector classifier (SVC) constructs a hyperplane or a set of hyperplanes in a high-dimensional space, which separates the training data into different classes. These are then used to classify a whole image, or series of images. The current standard SVM algorithm for classification used by the most popular mapping software (e.g., ENVI, EnMAP) is the C-SVC. The parameterization of a C-SVC strongly affects the final classification result. Yet, there is no rule of thumb to choose the optimal parameters when classifying satellite imagery. Optimal parameterization totally depends on the training data, and to determine it for a specific case, a time-consuming trial-and-error process is inevitable. In this work, advancements for the C-SVC algorithm are proposed to enhance its performance when used to classify remote sensing data, eliminating the need for a part of manual parametrization, while ensuring increasing its performance.