Mattias,
If the force is one-directional, I would just define contacts (Master-Slave or shared surface) only where compression is anticipated. Sad but true, you would better avoid using interfaces if there is such possibility. However, if you don't know exactly where tension occurs then you have to use interfaces. And yes, you are right, you need to define an interface with a low tensile resistance.
As to side surfaces, I would keep them not attached to concrete at all since they shouldn't contribute to the anchor resistance.
I wonder whether you have a 2D or 3D model? If your structure is similar along its width, I would recommend using 2D at least for model tuning. Later you can make a 3D and it would be way easier to define all boundary conditions when you have your 2D model working.
I made a sketch how I would model it myself: https://ibb.co/4RZjjHy
I would divide concrete into smaller pieces to have concrete as brick volumes (to mesh them with "Structured" type). Then each contact side can be defined separately: side surfaces may be not connected, pure compression can be master-slave or even interface (just for easier representation of contact stresses), top surface can be either not connected or interface in case you anticipate some compression at corners of anchor plate.
Interface definition is a separate topic which is not that simple (this is why I recommend avoiding them 
). This video can be helpful: https://www.youtube.com/watch?v=8HUcL_McfJo
