Couple options you can take

No matter how deep or how many tables in cluster, you can use two methods to get there:

Option 1:  (1) Create a standard table for each layer, in-dependant of each other; create a relationship table to contain their relationship, such as who owns whom (or who belongs to whom).

Option 2:  Create cluster tables - this can be two types:  (1) using index: for example, the first table contains header number (key) and descriptions, the second table contains header number (key) and item number (key) and other non-key info.  Now the relationship between header and item is established.  (2) using pointer:  for example, the first table contains header number (key), item table id, and descriptions.  Note:  item table id is not a real table but a pointer here, it points to the item table.

 

No matter what db tables, not much different.