This study explored the statistical classification problem, which dates back to the solution provided by the Fisher’s linear discriminant function. Multivariate discriminant models sometimes fail to achieve reasonable accuracy, especially when observations in some classes are difficult to be correctly classified. Todate, high prediction accuracy models are being sought upon, mainly for the non-separable problem. The support vector models have become more efficient as compared to the parametric models, in the last three decades, because of their robustness as they work through determining optimal linear hyperplanes. However, these models, which are used both for classification and regression are at times identified with significantly low levels of efficiency, requiring use of Kernels. Moreover, correct classification of patients with diabetes based on glycated haemoglobin (HbA1c) is an important indicator of blood glucose control and is a key requirement for level of management and medicinal intake compliance among patients diagnosed with diabetes. Thus, the need for models that account for the important associated demographic and related factors, which affect efficiency of diabetes control among patients. This study sought to achieve two-pronged aim, that is, to improve the performance of the support vector models by proposing hierarchical-based modelling and to predict diabetes’ disease control among patients using various datasets. The proposed hierarchical support vector machine (hSVM) and hierarchical support vector regression (hSVR) were based on the foundation of stratification and validated against the classical SVM through compairing the correct classification rates and mean square errors. Findings from the experimental studies all show that the modified hSVM and hSVR models significantly improve the correct classification rates across various data scenarios and minimize the mean square errors, respectively. Moreover, there are other advantages associated with proposed hSVM/hSVR such as scalability to large sample, besides giving detailed strata-based statistical analyses.