The paper presents an adaptive, hierarchical data fusion method for security biometric applications, with design elements for a software architecture. The methodology relies on a 2-layer (intra- and inter-modal) data fusion of texture (Haralick) features. The framework uses adaptive weighting based on the sample’s quality integrated into a quality-aware intra-modal attention (for the intra-modal layer). A gating mechanism for the inter-modal layer dynamically selects the fusion rule for the current context (combination of experts). The methodology could consider a security-by-design approach with cancellable templates. The score−/decision-level fusion can be used for the cases in which the feature representations are very heterogeneous or with missing features. This framework is suitable for systems working under varying conditions impacting on the data quality.