The paper presents a method for parallel speculative query execution support to be applied in relational database systems. The method is based on dynamic analysis of input query stream in databases serviced in SQLite. A special representation of queries in the form of multigraphs is employed. A middleware called the Speculative Layer is introduced which determines the most promising speculative queries for execution based on analysis of current stream of user queries. The proposed approach assumes a dynamic query speculation window organized in the query queue, which is converted into a query multigraph subdue to the analysis. The paper presents the query graph modelling method and the query graph analysis algorithms based on specific metrics. They aim in finding best candidate queries for speculative execution which fit currently accumulated query window. Experimental results are presented based on the proposed algorithms assessment using a real testbed database serviced in SQLite.