How to extract/build a Domain Model from Description Logic Knowledge Base(s)

Question

hopefully my question is not trivial or already answered elsewhere.

Problem Description
Let's assume we have a DL Knowledge Base K_mySys = (T_sys, A_mySys) representing a software architecture. The knowledge K_mySys models software components (e.g., APIs, Storages, DBs, Servers etc.) and their communication (e.g. Protocols, Authorizations, Permissions, Data flow etc.). The terminological part T_sys is shared amongst several possible models and the assertional part A_mySys is the instantiation of the resources constituting my specific software.
Now, let's also assume that we have a DL "Domain Knowledge" K_d = (T_d, A_d) that contains concepts and roles definitions, and possibly individuals thereof, of a specific domain. This "Domain Knowledge" can be thought of as extra knowledge that we want to combine with the model of our software to draft a report of some kind: for example, it might contain concepts related to security, access control, cryptography.
Some of the knowledge in K_mySys "intersects" K_d and some does not. The intersecting knowledge is relevant for the domain modeling, the remaining is not and can be dropped.

Question
Is there any procedure to:

1. Using the individuals from K_mySys, create a new ontology with assertions inferred from definitions in T_d ?
   
2. Exclude all the info not relevant for the domain d, that is, all individuals in K_mySys that do not belong to any concept in T_d?
   
3. Include all the info from T_d so that the final ontology is a sort of domain-specific model of the original model?

In Short: How to instantiate the domain knowledge K_d over existing individuals from a knowledge K_mySys?

I have googled extensively for papers on this problem and found results on Ontology Merging, Mapping, Matching, something on Projections and so on... but it doesn't seem to me that the formulation of the problem is the same as in my case. The problem vaguely resembles merging, only that I want to preserve the domain knowledge and drop the non-relevant system knowledge.

I am a little confused...
Is this a projection problem?
Is this a merging problem?
Can somebody please point me to relevant research?